JP4010063B2 - Wavelet decoding apparatus and method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention can be applied to a system for efficiently transmitting or storing images, and in particular, by inputting a bit stream encoded using wavelet transform encoding, resolution conversion of an arbitrary rational number multiple is possible. TECHNICAL FIELD The present invention relates to a wavelet decoding apparatus and method for realizing decoding accompanied with.
[0002]
[Prior art]
As a conventional typical image compression method, there is a JPEG (Joint Photographic Coding Experts Group) method standardized by ISO (International Organization for Standardization). This is a system that mainly compresses and encodes still image signals using DCT (discrete cosine transform), and is known to provide good encoded / decoded images when relatively high bits are allocated. . However, in the case of the DCT, if the number of encoded bits is reduced to some extent, block distortion peculiar to DCT becomes prominent and subjective deterioration becomes conspicuous.
[0003]
Apart from this, recently, research on a method of dividing an image signal into a plurality of bands using a filter that combines a high-pass filter and a low-pass filter called a filter bank and performing coding for each band has been actively conducted. It has become. Among them, wavelet coding is regarded as a promising new technology to replace DCT because it does not have the disadvantage that block distortion becomes remarkable due to high compression, which is a problem in DCT.
[0004]
Current products such as electronic still cameras and video movies use JPEG or MPEG (Moving Picture image coding Experts Group) as the image compression method and DCT as the conversion method. It is speculated that products that adopt this conversion method will appear on the market.
[0005]
[Problems to be solved by the invention]
However, studies for improving the efficiency of the coding scheme have been actively conducted by each research institution, but there are still few inventions aimed at specific commercialization utilizing the characteristics of the wavelet transform.
[0006]
Further, in the wavelet decoding accompanied with the conventional resolution conversion, the resolution can be reduced only by a power of 2 by its nature. This is due to the fact that the normal wavelet transform uses a two-part filter bank. Therefore, the synthesis filter bank in the decoding process can synthesize low-frequency components only with a power of 2. Therefore, the reduction rate of the decoded image is limited to 1 with a power of 2.
[0007]
On the other hand, when the resolution of the original image increases, it is considered that the demand for decoding at a resolution other than one that is a power of 2 increases. In other words, if it becomes possible to decode (decode) at an arbitrary rational number resolution including not only a power of 2 but also other than that, it will not be influenced by the constraints on the terminal side, It is thought that the use will spread.
[0008]
Therefore, the present invention has been made in view of such a situation, and an arbitrary rational number can be used for an image signal that has been compression-encoded using a wavelet transform as a conversion method without being influenced by restrictions on the terminal side. As a result, it is possible to efficiently store and display so-called thumbnail images that are frequently used in electronic still cameras, printers, etc., and images that have been converted in resolution (reduced images). It is an object of the present invention to provide a wavelet decoding apparatus and method that can be used and that can greatly expand the usage of various products.
[0009]
[Means for Solving the Problems]
The wavelet decoding apparatus and method according to the present invention entropy-decodes an encoded bit stream, sends quantization coefficients, dequantizes the quantization coefficients, sends transform coefficients, and scans the transform coefficients by a predetermined method. Then, the transform coefficients are rearranged, the rearranged transform coefficients are inversely transformed to generate a decoded image, and when performing wavelet inverse transform, the bandwidth of the transform coefficient of the high-frequency component is determined according to the resolution conversion magnification. As well as limiting Among the basic wavelet transform basic configuration composed of a plurality of upsamplers and synthesis filters, the high-frequency components at a predetermined level of the upsampling processing and synthesis filtering processing in the upsampler and synthesis filter for decoding on the high frequency side are omitted. The bandwidth restriction processing is performed on the high-frequency component of the bandwidth used in the inverse wavelet transform including the bandwidth of the reduction rate corresponding to the desired resolution conversion magnification, and the bandwidth corresponding to the reduction rate is extracted. Further, the image corresponding to the desired resolution conversion magnification is thinned out by down-sampling processing in the down-sampler, and the decoded image having a desired size is provided. This solves the above-mentioned problem.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
A preferred embodiment of the present invention will be described with reference to the drawings.
[0011]
The embodiment of the present invention is suitable for a system that efficiently transmits or stores images, and in particular, inputs a bit stream encoded using wavelet transform encoding, and performs resolution conversion of an arbitrary rational number multiple. The present invention realizes a wavelet decoding apparatus and method that realizes decoding accompanied with. Specific application examples include electronic cameras, portable / mobile image transmission / reception terminals (PDAs), printers, compression / decompression devices for satellite images, medical images, etc., or software modules thereof, games, and decompression of textures used in 3D CG. Or a software module thereof.
[0012]
FIG. 1 shows the overall configuration of a wavelet decoding apparatus according to an embodiment to which the wavelet decoding apparatus and method of the present invention are applied.
[0013]
The wavelet decoding apparatus according to the embodiment of the present invention shown in FIG. 1 includes an entropy decoding unit 1 that entropy-decodes an encoded bit stream 100, and an inverse unit that inversely quantizes a quantized coefficient 101 and sends a transform coefficient 102. The quantizing unit 2, the transform coefficient inverse scanning unit 3 that sends the transform coefficient 103 scanned by the transform coefficient 102 and rearranged by a predetermined method, the inverse transform of the rearranged transform coefficient 103, and the decoded image 104. A wavelet inverse transform unit 4 is provided.
[0014]
More specifically, the entropy decoding unit 1 performs predetermined entropy decoding on the encoded bitstream 100 transmitted from the wavelet encoding device or the encoding module. Here, as entropy decoding, generally used Huffman decoding or arithmetic decoding may be used. However, as a matter of course, it is necessary to perform a method corresponding to the entropy encoding process performed by the wavelet encoding device or the like.
[0015]
The inverse quantization unit 2 inversely quantizes the quantized coefficient 101 decoded and output by the entropy decoding unit 1 and outputs a transform coefficient 102. The inverse quantization unit 2 also needs to perform an operation that is integrated with the quantization processing performed by the wavelet encoding device.
[0016]
The transform coefficient inverse scanning unit 3 rearranges the transform coefficients 102 obtained by the inverse quantization unit 2 and outputs a new transform coefficient 103. The inverse scanning method here performs the inverse process of the scanning process performed by the wavelet encoding device.
[0017]
The wavelet inverse transform unit 4 inversely transforms the transform coefficient 103 to provide a final decoded image signal 104.
[0018]
Here, in the wavelet decoding apparatus according to the embodiment of the present invention, the wavelet inverse transform unit 4 is provided with band limiting means for adaptively limiting the band of the transform coefficient in accordance with a predetermined resolution transform magnification, and further an upsampler. The function of performing resolution conversion to an arbitrary rational number magnification is realized by adaptively arranging one or more of the downsampler and the synthesis filter.
[0019]
Before detailed description of the configuration and operation for the above-described arbitrary rational number resolution conversion processing in the wavelet decoding apparatus of the present embodiment, normal wavelet conversion processing and inverse wavelet conversion using FIGS. The configuration and operation for processing will be described below.
[0020]
FIG. 2 shows a basic configuration of a normal wavelet encoding apparatus.
[0021]
The wavelet encoding apparatus shown in FIG. 2 includes a wavelet transform unit 5, a transform coefficient scanning unit 6, a quantization unit 7, and an entropy coding unit 8 as its basic components.
[0022]
The wavelet transform unit 5 performs wavelet transform on the input image signal 105 and outputs the transform coefficient 106.
The transform coefficient scanning unit 6 rearranges the transform coefficients 106 from the wavelet transform unit 5 and outputs new transform coefficients 107. Note that the scanning in the transform coefficient inverse scanning unit 3 in FIG. 1 is a reordering process opposite to the scanning in the transform coefficient scanning unit 6.
[0023]
The quantization unit 7 quantizes the transform coefficient 107 supplied from the transform coefficient scanning unit 6 and outputs the quantized coefficient 108. The processing in the inverse quantization unit 2 in FIG. 1 is paired with the processing in the quantization unit 7.
[0024]
The entropy encoding unit 8 performs predetermined entropy encoding on the quantized coefficient 108 supplied from the quantization unit 7 and outputs the encoded bitstream 100. Here, as entropy coding, generally used Huffman coding or arithmetic coding may be used. The processing in the entropy decoding unit 1 in FIG. 1 is performed by the entropy coding unit 8. It corresponds to the processing of.
[0025]
FIG. 3 shows a configuration for performing normal wavelet transform processing. The configuration in FIG. 3 is a configuration example in the case where octave division, which is the most popular wavelet transform processing among several methods, is performed over a plurality of levels. In the case of FIG. 3, the number of levels is 3 (level 1 to level 3), and the image signal is divided into a low band and a high band, and only the low band component is divided hierarchically. In FIG. 3, for the sake of convenience, wavelet transform processing for a one-dimensional signal (for example, a horizontal component of an image) is given as an example. it can.
[0026]
In FIG. 3, the input image signal 105 is first band-divided by the analysis low-pass filter 81 and the analysis high-pass filter 82, and the obtained low-frequency side signal 113 and high-frequency side signal 119 are respectively down-corresponding. The resolution is reduced by half by the samplers 83 and 84 (level 1).
[0027]
Of the outputs from the down samplers 83 and 84, the low-frequency signal 114 is further divided into bands by the analysis low-pass filter 85 and the analysis high-pass filter 86. The resolutions of the signals 116 and 115 obtained by the band division are further reduced by half by the down samplers 87 and 88 (level 2).
[0028]
Of the outputs from the down samplers 87 and 88, the low-frequency signal 117 is further divided into bands by the analysis low-pass filter 89 and the analysis high-pass filter 90. The resolutions of the signals 119 and 118 obtained by the band division are further reduced by half by the downsamplers 91 and 92 (level 3).
[0029]
By performing such processing up to a predetermined level, signals in each band obtained by hierarchically dividing the low-frequency signal into bands are sequentially generated. In the example of FIG. 3, it is shown that the LLL signal 109, the LLH signal 110, the LH signal 111, and the H signal 112 are generated as a result of band division to level 3. Note that L in the LLL signal 109 and the LLH signal 110 represents a low-frequency component, and H represents a high-frequency component.
[0030]
FIG. 4 illustrates band components obtained as a result of band division of a two-dimensional image up to level 2. However, the notation of L and H in FIG. 4 is different from that in FIG. 3 in which one-dimensional signals are handled. Note that LL in FIG. 4 means that the horizontal and vertical components are both L (low frequency), and LH means that the horizontal component is H (high frequency) and the vertical component is L (low frequency). . In the figure, X_SIZE means the resolution in the vertical direction (X direction), and Y_SIZE means the resolution in the horizontal direction (Y direction).
[0031]
That is, in FIG. 4, the two-dimensional original image is first divided into four components LL, LH, HL, and HH by level 1 band division (horizontal and vertical directions), and then the LL component is divided into level 2 bands. By the division (horizontal / vertical direction), it is further divided into four components LLLL, LLHL, LLLH, and LLHH.
[0032]
FIG. 5 shows an image example when the band division of FIG. 4 is applied to an actual image. From FIG. 5, it is understood that the image includes most of the information in the low-frequency component. Recognize.
[0033]
Next, FIG. 6 shows a configuration for performing a normal wavelet inverse transform process without performing a resolution conversion operation.
[0034]
Of the band components (LLL signal 109, LLH signal 110, LH signal 111, and H signal 112) that are the outputs of the wavelet transform unit described in FIG. 3, the LLL signal 109 and the LLH signal 110 are the upsamplers 9 and 11, respectively. To upsample to twice the resolution.
[0035]
The signal generated by up-sampling the LLL signal 109 by the up-sampler 9 is generated by the synthesis low-pass filter 10, and the signal generated by up-sampling the LLH signal 110 by the up-sampler 11 is generated by the synthesis high-pass filter. 12 are respectively filtered and sent to the adder 13.
[0036]
The adder 13 performs band synthesis on both signals. By the processing so far, the level 3 inverse transformation is completed.
[0037]
Similarly, the above-described processing is repeated up to level 1 to output the final decoded image 104 after inverse transformation.
[0038]
That is, the output signal of the adder 13 is further upsampled to a double resolution by the upsampler 14, filtered by the synthesis low-pass filter 15, and sent to the adder 18.
[0039]
The LH signal 111 is upsampled to a double resolution by the upsampler 16, filtered by the synthesis high-pass filter 17, and sent to the adder 18.
[0040]
The adder 18 performs band synthesis on both signals from the synthesis low-pass filter 15 and the synthesis high-pass filter 17. By the processing so far, the inverse transformation of level 2 is completed.
[0041]
The output signal of the adder 18 is further upsampled to a double resolution by the upsampler 19, filtered by the synthesis low-pass filter 20, and sent to the adder 23.
[0042]
The H signal 112 is upsampled to double the resolution by the upsampler 21, filtered by the synthesis high-pass filter 22, and sent to the adder 23.
[0043]
The adder 23 band-synthesizes both signals from the synthesis low-pass filter 20 and the synthesis high-pass filter 22. By the processing so far, the level 1 inverse transformation is completed.
[0044]
The above is the basic configuration and basic operation of normal wavelet transform processing and wavelet inverse transform processing.
[0045]
Incidentally, in FIG. 6, the LLL signal 109 itself corresponds to a reduced image of 1/8 of the original image. Further, the up-sampler 9 up-samples the signal and the low-frequency side signal 120 of level 3 that has passed through the synthesis low-pass filter 10 is also up-sampled by the up-sampler 11 twice. Further, the synthesis high-pass filter A signal 122 obtained by band synthesis of the high-frequency signal 121 of level 3 that has passed through 12 by the adder 13 corresponds to a reduced image that is a quarter of the original image. Similarly, the output signal 123 of the level 2 adder 18 corresponds to a reduced image that is a half of the original image.
[0046]
Therefore, if the LLL signal 109 and the signals 122 and 123 are extracted, a reduced image having a power of 2 can be generated. Note that a reduced image having a power-of-two power of 2 can be generated by the same method even if the number of levels is other.
[0047]
FIG. 7 shows the band division characteristics of the input signals (LLL signal 109, LLH signal 110, LH signal 111, H signal 112) of FIG. Here, since the present invention is intended for digital signals, in FIG. 7, the horizontal axis indicates a low frequency component as it approaches 0, 2π, and a high frequency component as it approaches π. Further, each reduced image corresponding to each signal 109, 122, 123 in FIG. 6 and the decoded image signal 104 of the original resolution each have a bandwidth as shown in FIG. It can be seen that the reduction rate of the image and the bandwidth occupancy rate (bandwidth occupancy rate) in each image coincide.
[0048]
Therefore, for example, in order to perform reduction conversion of an arbitrary rational number, it is necessary to extract a bandwidth corresponding to the reduction rate. However, as can be seen from the band division characteristics shown in FIG. 7, the wavelet transform / inverse transform can synthesize the bandwidth only with a band occupancy ratio of 1 that is a power of 2.
[0049]
For this reason, the wavelet inverse transform unit 4 of the wavelet decoding apparatus shown in FIG. 1 determines a band to be limited based on the reduction rate, and is given by the reduction rate in the determination. By minimizing the difference between the effective bandwidth and the bandwidth used for inverse wavelet transformation, it is possible to achieve bandwidth extraction that matches the reduction ratio as described above, and reduction conversion that is an arbitrary rational multiple. ing.
[0050]
In the first specific example of the embodiment of the present invention, an inverse wavelet transform accompanied by a reduction conversion of 1/3 will be described as an example of arbitrary rational multiple resolution conversion.
[0051]
As described above, when the reduction ratio is 1/3, the bandwidth occupancy must also be 1/3. That is, as shown in FIG. 8, it is necessary to extract the bandwidth of the band 128. In this case, first, band limitation processing is performed on the LH signal 111 to extract the band 124 (shaded portion) in FIG. 8 and combine it with the band 122 to extract the third band 128. It becomes.
[0052]
FIG. 9 shows a schematic configuration of the wavelet inverse transform unit 4 with a 1/3 reduction transform for realizing the above process as the first specific example. In the configuration of FIG. 9, the same reference numerals as those in FIG. 6 are assigned to the same components as those in FIG. 9 is a path provided in the normal wavelet decoding apparatus shown in FIG. 6, but is omitted in the wavelet decoding apparatus of the present embodiment. Represents the route that was made.
[0053]
That is, in the wavelet inverse transform unit 4 of the first specific example shown in FIG. 9, the LLL signal 109 and the LLH signal 110 are upsampled to double the resolution by the upsamplers 9 and 11, respectively, and the corresponding synthesis low-pass After filtering by the filter 10 and the synthesizing high-pass filter 12, the adder 13 band-synthesizes both signals. The signal 122 (the signal in the band 122 in FIG. 8) obtained by the band synthesis in the adder 13 corresponds to a reduced image that is a quarter of the original image. Is completed.
[0054]
The output signal 122 of the adder 13 is further upsampled to a double resolution by the upsampler 14, filtered by the synthesis low-pass filter 15, and sent to the adder 18.
[0055]
Further, the bandwidth of the LH signal 111 is limited by the bandwidth limiter 30 as will be described later. The band-limited signal 124 (the signal in the band 124 shown in FIG. 8) is up-sampled by the up-sampler 16 to double the resolution. The up-sampled signal 127 is further filtered by the synthesis high-pass filter 17 and sent to the adder 18 as a signal 125.
[0056]
The adder 18 performs band synthesis on the signals 126 and 125 from the synthesis low-pass filter 15 and the synthesis high-pass filter 17. The signal 128 obtained by the band synthesis in the adder 18 (the signal in the band 128 in FIG. 8) is further upsampled to a double resolution by the upsampler 19 and then filtered by the synthesis low pass filter 20. Is done. The output signal 129 from the synthesis low-pass filter 20 corresponds to an image having the same resolution as the original image.
[0057]
In this first specific example, the signal 129 from the low-pass filter 20 is further reduced to a third resolution by the down sampler 31 in order to finally generate an image decoded to a third resolution. Down-sampled (thinning-out process). As a result, a decoded image signal 130 corresponding to one-third reduced image is obtained. In addition, when only a third band signal is required as in the first specific example, the H signal 112 is not required. Therefore, the processing of the double upsampler and the synthesis high-pass filter that receive the H signal 112 as shown in FIG. 6 is unnecessary, and the amount of calculation can be reduced.
[0058]
Here, the band limiting unit 30 in the case of the first specific example is placed in the preceding stage of the upsampler 16 and the synthesis filter 17 and performs the filtering process of the conversion coefficient of the LH signal 111 (high frequency component). Since the LH signal 111 is down-sampled by the wavelet encoding device, it must be designed as a filter corresponding to the band at the resolution.
[0059]
FIG. 10 shows a frequency band state in each generation process when the LH signal 111 is generated in the wavelet transform unit shown in FIG. Note that the numbers in parentheses in the figure correspond to the signals shown in each figure described above, and the hatched part in the figure indicates the band 124 to be extracted. Moreover, ω on the horizontal axis in the figure means a normalized angular frequency.
[0060]
In FIG. 10, the band included in the input image signal 105 is indicated by a state 131. The band of the signal 113 obtained by processing the input image signal 105 by the low-pass filter 81 for analysis is in a state 132 because the high band is cut, and the band of the signal 114 is halved by the down sampler 83. Since it is down-sampled twice, the band is in the state 133. Further, the band of the signal 115 is the band as in the state 134 because the low band side is cut by the processing by the analysis high pass filter 86, and the downsampler 88 performs half-down downsampling. Thus, the LH signal 111 is generated. The LH signal 111 also has a band like the state 135 due to the influence of downsampling. In the state 135, the band shifts to the 2π side and a folding component toward the low band side is generated, so that it can be regarded as the state 136 as shown in FIG. Therefore, in the resolution of the LH signal 111, the band 124 in FIG. 8 corresponds to the band 137 in FIG. For this reason, when the band limiting unit 30 is arranged for the LH signal 111 as shown in FIG. 9, a high-pass filter having a filter characteristic (amplitude) as shown in FIG. 12 is used as the band limiting unit 30. .
[0061]
The above is the configuration and operation of the wavelet inverse transform unit 4 when the reduction ratio is 1/3 as the first specific example. According to the wavelet inverse transform unit 4 of the first specific example, the bandwidth in the wavelet transform region is reduced to one third by the above-described processing, so that the occurrence of blur due to noise such as aliasing and lack of high-frequency components is prevented. be able to. In other words, a high-precision extraction method of the effective band given from the reduction ratio produces an effect of improving the quality of the decoded image.
[0062]
Next, a wavelet decoding apparatus that performs 1/5 reduction conversion will be described as a second specific example of the embodiment of the present invention.
[0063]
As in the first specific example, when the reduction ratio is 1/5, the bandwidth occupation ratio must also be 1/5. For this purpose, it is necessary to extract the bandwidth of the band 140 as shown in FIG. In this case, band limitation processing is performed on the LLH signal 110 to extract the band 138 (shaded portion), and by combining this with the band of the LLL signal 109, the band 140 in FIG. 13 can be extracted.
[0064]
FIG. 14 shows a schematic configuration of the wavelet inverse transform unit 4 when the wavelet decoding apparatus of FIG. 1 performs decoding with the resolution reduced to 1/5 as the second specific example. In the configuration of FIG. 14, the same reference numerals as those of the respective drawings are assigned to the same components as those of the respective drawings. Further, each high-frequency path indicated by a dotted line in FIG. 14 is a path provided in the normal wavelet decoding apparatus shown in FIG. 6, but in the wavelet decoding apparatus of the present embodiment, Represents an abbreviated route.
[0065]
That is, in the wavelet inverse transform unit 4 of the second specific example shown in FIG. 14, the LLL signal 109 is upsampled to double the resolution by the upsampler 9 and further filtered by the synthesis low pass filter 10, and then the adder 13 Sent to.
[0066]
On the other hand, the bandwidth of the LLH signal 110 that is a high frequency component of level 3 is limited by the band limiting unit 32 as described later. The band-limited signal 138 (the signal in the band 138 in FIG. 13) is upsampled to a double resolution by the upsampler 10 and further filtered by the synthesis low-pass filter 10. The filtered signal 139 is sent to the adder 13.
[0067]
The output signal 140 of the adder 13 (the signal in the band 140 in FIG. 13) is further upsampled to a double resolution by the upsampler 14 and then filtered by the synthesis low-pass filter 15.
[0068]
The filtered signal 141 is further upsampled to a double resolution by the upsampler 19 and then filtered by the synthesis low-pass filter 20. The output signal 129 from the synthesis low-pass filter 20 is an image having the same resolution as the original image.
[0069]
In this second specific example, the output signal 129 is further downsampled (decimated by the downsampler 33) to 1/5 resolution in order to finally generate an image decoded to 1/5 resolution. It is processed. As a result, a decoded image signal 142 of a reduced image of 1/5 is obtained. Thus, when only one fifth of the band component is required, the LH signal 111 and the H signal 112 are not required. Therefore, as shown in FIG. 6, the processing of the double upsampler and the synthesis high-pass filter that receive the LH signal 111 and the H signal 112 as input becomes unnecessary, and the amount of calculation can be reduced.
[0070]
Here, the band limiting unit 32 in the case of the second specific example is placed in the preceding stage of the upsampler 10 and the synthesis filter 12 and performs the filtering process of the conversion coefficient of the LLH signal 110 (high frequency component). Since the LLH signal 110 is down-sampled by the wavelet encoding device, it needs to be designed as a filter corresponding to the band at the resolution.
[0071]
FIG. 15 shows the state of the frequency band in each generation process when the LLH signal 110 is generated in the wavelet transform unit of FIG. Note that the numbers in parentheses in the figure correspond to the signals shown in each figure described above, and the hatched portion in the figure indicates the band 138 to be extracted. Moreover, ω on the horizontal axis in the figure means a normalized angular frequency. Further, the band of the state 143 corresponds to the band of the signal 114, and the frequency band until the signal 114 is generated is already shown in the state 131 to the state 133 in FIG. 10 (except for the hatched portion). Has been.
[0072]
In FIG. 15, in the processing after the output of the signal 114, a signal 116 whose high frequency is cut by the analysis low-pass filter 85 is generated, and the band component of the signal 116 becomes a state 144. Further, the signal 117 is generated by the half-down sampling in the down sampler 87, and the band of the signal 117 is represented by a state 145. Next, the processing at level 3 is performed, and the signal 117 is band-limited by the analysis high-pass filter 82 to generate the signal 118. Since this signal 118 is cut off in the low frequency band, the band becomes a state 146 in FIG. Finally, half down-sampling by the down sampler 92 is performed, and the LLH signal 110 is generated. The bandwidth of LLH signal 110 is represented by state 147. In this state 147, since the band shifts to the 2π side and a folding component to the low band side is generated, it can be regarded as a state 148 as shown in FIG. Therefore, in the resolution of the LLH signal 110, the band 138 in FIG. 13 corresponds to the band 149 in FIG. For this reason, when the band limiting unit 32 is arranged for the LLH signal 110 as shown in FIG. 14, a high-pass filter having the filter characteristic (amplitude) shown in FIG. 17 is used as the band limiting unit 32.
[0073]
The above is the configuration and operation of the wavelet inverse transform unit 4 when the reduction ratio is 1/5 as the second specific example. According to the wavelet inverse transform unit 4 of the second specific example, the bandwidth in the wavelet transform region is reduced to 1/5 by the above-described processing, so that the occurrence of blur due to noise such as aliasing or lack of high frequency components is prevented. be able to. In other words, a high-precision extraction method of the effective band given from the reduction ratio produces an effect of improving the quality of the decoded image.
[0074]
Next, a wavelet decoding apparatus that performs two-thirds reduction conversion will be described as a third specific example of the embodiment of the present invention.
[0075]
As in the first and second specific examples, when the reduction ratio is two thirds, the band occupation ratio must be two thirds. For this purpose, it is necessary to extract the bandwidth of the band 153 as shown in FIG. In this case, band limitation processing is performed on the H signal 112 to extract the band 150 (shaded portion), and this and the band of the signal 123 are combined to extract the band 153 of two-thirds in FIG. It becomes possible.
[0076]
FIG. 19 shows a schematic configuration of the wavelet inverse transform unit 4 when the wavelet decoding apparatus in FIG. 1 performs decoding with the resolution reduced to two-thirds as the third specific example. In the configuration of FIG. 19, the same reference numerals as those of the respective drawings are attached to the same components as those of the respective drawings. Further, each high frequency path indicated by a dotted line in FIG. 19 is a path provided in the normal wavelet decoding apparatus shown in FIG. 6, but the wavelet decoding apparatus according to the present embodiment. Represents a route that cannot be provided.
[0077]
That is, in the wavelet inverse transform unit 4 of the third specific example shown in FIG. 19, the LLL signal 109 and the LLH signal 110 are up-sampled to double the resolution by the up-samplers 9 and 11, respectively, After being respectively filtered by 12, an adder 13 performs band synthesis for both. The signal 122 band-combined by the adder 13 corresponds to a reduced image of a quarter of the original image, and the inverse transformation of level 3 is completed by the processing so far.
[0078]
The output signal 122 of the adder 13 is further upsampled to a double resolution by the upsampler 14, filtered by the synthesis low-pass filter 15, and sent to the adder 18.
[0079]
The LH signal 111 is upsampled to a double resolution by the upsampler 16, filtered by the synthesis high-pass filter 17, and sent to the adder 18.
[0080]
The adder 18 performs band synthesis of signals from the synthesis low-pass filter 15 and the synthesis high-pass filter 17. By the processing so far, the inverse transformation of level 2 is completed. The signal 123 after band synthesis by the adder 18 is further upsampled to a double resolution by the upsampler 19, filtered by the synthesis low-pass filter 20, and sent to the adder 23.
[0081]
Further, the bandwidth of the H signal 112 is limited by the band limiting unit 34 as will be described later. The band-limited signal 150 (the signal in the band 150 in FIG. 18) is upsampled to a double resolution by the upsampler 21 and further filtered by the synthesis high-pass filter 22. The filtered signal 151 is sent to the adder 23.
[0082]
The adder 23 performs band synthesis of signals from the synthesis low-pass filter 20 and the synthesis high-pass filter 22. By the processing so far, the level 1 inverse transformation is completed.
[0083]
In this third example, the signal 153 is doubled by the upsampler 24 (twice the resolution of the original image) in order to finally generate an image decoded to two-thirds resolution. After being up-sampled, the signal is filtered by the synthesis low-pass filter 25, and the signal 154 is further down-sampled (decimation processing) by the down-sampler 31 to a resolution of 1/3. As a result, a decoded image signal 142 of a reduced image of 2/3 is obtained. Even if the inverse transformation up to level 0 is performed in this way, in the third specific example, a double upsampler and a synthesis high-pass filter are unnecessary on the high-frequency side of level 0.
[0084]
Here, the band limiting unit 34 in the case of the third specific example is placed in the preceding stage of the upsampler 21 and the synthesis filter 22, and performs filtering processing of the conversion coefficient of the level 1 H signal 112 (high frequency component). However, since the H signal 112 is down-sampled by the wavelet encoding device, it is necessary to design it as a filter corresponding to the band at the resolution.
[0085]
FIG. 20 shows a frequency band state in each generation process when the H signal 112 is generated in the wavelet transform unit of FIG. The numbers in parentheses in the figure correspond to the signal components shown in each figure described above, and the hatched part in the figure indicates the band 150 to be extracted. Moreover, ω on the horizontal axis in the figure means a normalized angular frequency.
[0086]
In FIG. 20, the band included in the input image signal 105 is indicated by a state 156. Next, the band of the signal 119 that has been subjected to the analysis high-pass filter 82 processing for the state 105 is changed to the state 157 because the low frequency band is cut, and the downsampler 84 performs a half-sampling of downsampling. The band of the signal 112 that has been subjected to is a band like the state 158. In the state 158, since the band is shifted to the 2π side and a folded component toward the low band side is generated, the band distribution can be regarded as the state 159. Accordingly, the band 150 in FIG. 18 corresponds to the band 160 in FIG. 20 in the resolution of the H signal 112. Therefore, when the band limiting unit 34 is arranged for the H signal 112 as shown in FIG. 19, the high-pass filter having the filter characteristics (amplitude) shown in FIG. 12 is used as the band limiting unit 34.
[0087]
The above is the configuration and operation of the wavelet inverse transform unit 4 when the reduction ratio is 2/3 as the third specific example. According to the wavelet inverse transform unit 4 of the third specific example, the bandwidth in the wavelet transform region is reduced to two-thirds by the above-described processing, thereby preventing the occurrence of blur due to noise such as aliasing and lack of high-frequency components. be able to. In other words, a high-precision extraction method of the effective band given from the reduction ratio produces an effect of improving the quality of the decoded image.
[0088]
Next, as a fourth specific example of the embodiment of the present invention, a wavelet decoding apparatus that performs one-third reduction conversion as in the first specific example will be described.
[0089]
FIG. 21 shows a schematic configuration of the wavelet inverse transform unit 4 when the wavelet decoding apparatus of FIG. 1 performs decoding with the resolution reduced to one-third as the fourth specific example. In the configuration of FIG. 21, the same reference numerals as those of the respective drawings are attached to the same components as those of the respective drawings. Further, when the resolution is reduced to one third, the band occupancy rate is also reduced to one third as in the first specific example.
Here, the configuration of the fourth specific example shown in FIG. 21 is the same as that of the first specific example except that the band limiting unit of FIG. 9 and the arrangement of the upsampler 16 and the synthesis high-pass filter 17 are replaced. It becomes the same structure as FIG.
[0090]
That is, in FIG. 21, the LH signal 1111 is up-sampled by the up-sampler 16 to double the resolution. The up-sampled signal 161 is filtered by the synthesis high-pass filter 17 and sent to the band-pass valve unit 36 as a signal 162.
[0091]
In the case of the fourth specific example, the signal 162 subject to band limitation is doubled in resolution by the upsampler 16 by the upsampler 16 and the synthesis high-pass filter 17 to the LH signal 111. Yes. Therefore, the band limiting unit 36 needs to design a filter based on the band at this resolution.
[0092]
FIG. 22 shows the state of the frequency band in each process of generating the signal 162 from the LH signal 111 by the processing of double upsampling and high-pass filter (synthesis). The numbers in parentheses in the figure correspond to the signals shown in each figure described above, and ω on the horizontal axis means the normalized angular frequency.
[0093]
In FIG. 22, the state 163 indicates the same band of the signal 111 as that in the state 136 in FIG. 11, and the hatched portion in the drawing is the band of the signal 125 to be extracted. The signal 161 obtained by up-sampling the signal 111 twice has a band distribution as shown in the state 164. A band indicated by a dotted line in the figure indicates an imaging component (a spectrum component newly generated by upsampling). Further, the signal 162 of the signal 162 obtained by performing high-pass filtering on the signal 161 has its imaging component cut, and its band has a distribution as indicated by a state 165.
[0094]
From the above, in the case of the fourth specific example, in order to extract the band of the signal 125 in FIG. 21, a bandpass filter having a filter characteristic (amplitude) as shown in FIG. .
[0095]
As described above, in the embodiment of the present invention, it is possible to realize wavelet decoding with resolution conversion at an arbitrary rational number magnification. That is, for example, in the case of reduction, there is an effect of reducing the amount of calculation by omitting the high frequency component below the level subject to band limitation from the decoding process. Therefore, it leads to cost reduction when hardware is used. In addition, the high-accuracy extraction method of the effective band given by the reduction ratio prevents noise such as aliasing and blur due to missing high-frequency components, and produces an effect of improving the quality of the decoded image.
[0096]
Further, according to the present embodiment, there is no restriction condition on the wavelet encoding device side. Accordingly, there is an effect that a wavelet decoded image accompanied by resolution conversion of an arbitrary rational number can be obtained by inputting an encoded bit stream generated by a normal most general wavelet transform and wavelet encoding device.
[0097]
【The invention's effect】
In the wavelet decoding apparatus and method of the present invention, the band limitation of the transform coefficient is performed according to the resolution transform magnification in the case of inverse wavelet transform, and upsampling, downsampling, and synthesis filtering are performed according to a predetermined resolution transform magnification. By adaptively performing the above, it is possible to decode (decode) an image signal that has been compression-encoded using wavelet transform as the conversion method, with an arbitrary rational number resolution, regardless of the restrictions on the terminal side. As a result, for example, so-called thumbnail images frequently used in electronic still cameras, printers, etc., and the resolution-converted images (reduced or enlarged images) of the original images can be efficiently stored and displayed for use in various products. Applications can be greatly expanded.
[0098]
That is, according to the present invention, the band image stored in the image memory can be displayed on the screen as a thumbnail image or a reduced image as necessary, so that the process of generating the band-divided image and the process of encoding are shared. Thus, there is an effect that the processing efficiency can be realized. Therefore, there is no need for a circuit for generating a thumbnail image or the like, so there is an effect of reducing the hardware scale. Further, for example, by adding an external storage medium to the apparatus of the present invention and storing / holding the encoded bit stream therein, it is possible to store / hold the encoded bit stream of many images in the external storage medium. . In addition, since it is not necessary to always store and hold thumbnail images or reduced images in the image memory, the encoded bit stream of the thumbnail image or reduced image to be viewed is read from an external storage medium at any time, decoded and displayed on the screen. Therefore, the use efficiency can be improved.
[Brief description of the drawings]
FIG. 1 is a block circuit diagram showing an overall configuration of a wavelet decoding apparatus according to an embodiment of the present invention.
FIG. 2 is a block circuit diagram showing an overall configuration of a wavelet encoding device corresponding to the wavelet decoding device according to the embodiment of the present invention;
FIG. 3 is a block circuit diagram showing a basic configuration (up to level 3) of a normal wavelet transform unit;
FIG. 4 is a diagram illustrating band division (division level = 2) of a two-dimensional image.
FIG. 5 is a diagram illustrating each band image when band division (division level = 2) is performed on an actual image.
FIG. 6 is a block circuit diagram showing a basic configuration (up to level 3) of a normal wavelet inverse transform unit;
FIG. 7 is a diagram showing band division characteristics (octave division).
FIG. 8 is a diagram illustrating a band division characteristic to be processed at the time of resolution conversion of 1/3.
FIG. 9 is a block circuit diagram showing a configuration of a wavelet inverse transform unit with a 1/3 reduction resolution conversion as a first specific example;
FIG. 10 is a diagram illustrating a frequency band in the process of generating an LH signal by a wavelet transform unit.
FIG. 11 is a diagram illustrating a frequency band in the resolution of an LH signal.
FIG. 12 is a diagram showing amplitude characteristics of a high-pass filter used in the band limiting unit of the first specific example.
FIG. 13 is a diagram illustrating a band division characteristic to be processed when a resolution conversion of 1/5 is performed.
FIG. 14 is a block circuit diagram showing a configuration of a wavelet inverse transform unit with a 1/5 reduction resolution conversion as a second specific example;
FIG. 15 is a diagram illustrating a frequency band in a process of generating an LLH signal by a wavelet transform unit.
FIG. 16 is a diagram illustrating a frequency band in the resolution of an LLH signal.
FIG. 17 is a diagram illustrating amplitude characteristics of a high-pass filter used in the band limiting unit of the second specific example.
FIG. 18 is a diagram illustrating a band division characteristic to be processed at the time of resolution conversion of 2/3 times.
FIG. 19 is a block circuit diagram showing a configuration of a wavelet inverse transform unit with a reduced resolution conversion of 2/3 as a third specific example;
FIG. 20 is a diagram illustrating a frequency band in the process of generating an H signal by a wavelet transform unit.
FIG. 21 is a block circuit diagram showing a configuration of a wavelet inverse transform unit accompanied by a reduced resolution conversion of 1/3 as a fourth specific example;
FIG. 22 is a diagram showing frequency bands in the process of performing up-sampling and high-pass filter processing on an LH signal.
FIG. 23 is a diagram illustrating amplitude characteristics of a bandpass filter used in the band limiting unit of the third specific example.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Entropy decoding part, 2 Inverse quantization part, 3 Transformation coefficient inverse scanning part, 4 Wavelet inverse transformation part, 5 Wavelet transformation part, 6 Transformation coefficient scanning part, 7 Quantization part, 8 Entropy encoding part, 9 2 times Up sampler, 10, 15, 20, 25 low pass filter for synthesis, 12, 17, 22 high pass filter for synthesis, 12, 18, 12 adder, 81, 85, 89 low pass filter for analysis, 82, 86, 90 analysis High-pass filter, 83, 84, 87, 88, 91, 92 1/2 downsampler, 31/3 downsampler, 33/5 downsampler, 30, 32, 34, 36 Bandwidth limiter

Claims (2)

Entropy decoding means for entropy decoding the encoded bitstream and sending quantized coefficients;
Inverse quantization means for inversely quantizing the quantization coefficient and sending a transform coefficient;
A transform coefficient inverse scanning means for scanning the transform coefficients by a predetermined method and rearranging the transform coefficients;
Wavelet inverse transform means for inversely transforming the rearranged transform coefficients to provide a decoded image;
The wavelet inverse transform means has band limiting means for restricting the bandwidth of the conversion coefficient of the high frequency component according to the resolution conversion magnification, and among the wavelet inverse transform basic configuration composed of a plurality of upsamplers and synthesis filters, A wavelet inverse including a band of a reduction ratio corresponding to a desired resolution conversion magnification while omitting a predetermined level of the high-frequency component of up-sampling processing and synthesis filtering processing in an up-sampler and synthesis filter for decoding on the high frequency side The band limiting process by the band limiting unit is performed on the high band component of the bandwidth used in the conversion to extract a bandwidth corresponding to the reduction ratio, and the desired resolution conversion is performed by the down sampling process in the down sampler. by thinning out image according to the magnification, this subjecting the decoded image of the desired size Wavelet decoding apparatus according to claim. Entropy decodes the encoded bitstream and sends the quantized coefficients,
Dequantize the quantized coefficient and send the transform coefficient,
Scan the conversion coefficients in a predetermined way to rearrange the conversion coefficients,
Inversely transform the rearranged transform coefficients to generate a decoded image,
When performing the wavelet inverse transform, the bandwidth of the high frequency component conversion coefficient is limited according to the resolution conversion magnification, and the high frequency side of the wavelet inverse transform basic configuration composed of a plurality of upsamplers and synthesis filters In the case of inverse wavelet transform including a band of a reduction ratio corresponding to a desired resolution conversion magnification while omitting a predetermined level of the high-frequency component in the upsampling process and synthesis filtering process in the upsampler and synthesis filter for decoding The bandwidth corresponding to the reduction ratio is extracted by performing the bandwidth restriction process on the high-frequency component of the bandwidth used for the image, and further the image corresponding to the desired resolution conversion magnification by the downsampling process in the downsampler thinning out, characterized in that subjecting the decoded image of the desired size Webure Door decoding method.

Images